{ "schema_version": "1.0", "page_kind": "article", "pack_kind": "generate", "slug": "ask-about-this-page", "title": "Ask About This Page", "dek": "Every article on this hub has a room now, where a model on our own machines answers from that page’s own words and leaves its first draft on the page, marked where the checks took something out. This is how it is built, what it keeps, what it refuses, and what it got wrong on the way here.", "published": "2026-09-16", "series": [ "notes" ], "licence": "CC BY 4.0", "status": "pending-judge", "pack_note": "no judge is seated, so every chip is held", "notice": "items with published:false are held and are not this page's published words", "source": { "url": "https://research.strata2signal.com/ask-about-this-page/", "md_url": "https://research.strata2signal.com/ask-about-this-page/index.md", "md_sha": "ec55861c99e9c0b408a15bf2b674b9cac04d533b19036ab8a4abb82b57b93991", "html_sha": "a8db0b69c060d7b5fa93048ace2e550524fe7bce0fb02bb7fd5a52ed358af907", "anchors_sha": "8fa6fa104b1db4e88fc98869462fb648d3bf9e1170b4f57125a5232f02e825ae" }, "short": { "paragraph": "Every article on this hub now has a room where a model on our own machines answers from that page's own words. The buttons cost no model call; free text wakes one seat with the whole article in its window, no retrieval. Six checks run over the draft in public and the marks are laid on sentence by sentence before anything is called an answer; on the launch run they struck six sentences of 569, every one on a question the page could not answer. It passed a 62-question exam on the version that went live, and this page says what it still cannot measure.", "from_dek": false, "counts": { "words": 4819, "minutes": 22, "tables": 1, "kit": true }, "bullets": [] }, "sections": [ { "id": "the-room-behind-every-article-and-what-it-opens", "heading": "The room behind every article, and what it opens", "level": 2, "span": [ 1115, 3245 ], "chunks": [ [ 1115, 3245 ] ], "chars": 2130, "digest": "every article has a room with a question box and buttons. the room reads the article and a model drafts an answer subject to six checks. forty-six articles have a room. the room reads the article whole and runs on a machine on this workshop's own network.", "digest_skipped": null }, { "id": "five-words-this-page-leans-on", "heading": "Five words this page leans on", "level": 2, "span": [ 3245, 4301 ], "chunks": [ [ 3245, 4301 ] ], "chars": 1056, "digest": "the page relies on five terms: room, seat, the file beside the article, fence, and abstain. a seat is a gemma-class open-weight writer on a vLLM runtime with FP8 weights. a fence is a marked-off block in the prompt that the model reads as text.", "digest_skipped": null }, { "id": "where-the-answers-come-from", "heading": "Where the answers come from", "level": 2, "span": [ 4301, 6277 ], "chunks": [ [ 4301, 6277 ] ], "chars": 1976, "digest": "buttons use the author's own words from the file beside the article, costing no model. free text wakes a model that reads the whole article inside a fence. the thread lives in the page itself, and no cloud model runs anywhere in this chain.", "digest_skipped": null }, { "id": "what-happens-to-a-draft-before-you-see-it", "heading": "What happens to a draft before you see it", "level": 2, "span": [ 6277, 10279 ], "chunks": [ [ 6277, 10279 ] ], "chars": 4002, "digest": "a model's draft streams onto the page and undergoes six checks. these checks verify figures, citations, names, section markers, and redactions. sentences that fail are struck through with a reason. the final answer appears below the draft with numbered footnotes to the sections.", "digest_skipped": null }, { "id": "what-it-keeps-and-what-it-never-writes-down", "heading": "What it keeps, and what it never writes down", "level": 2, "span": [ 10279, 11598 ], "chunks": [ [ 10279, 11598 ] ], "chars": 1319, "digest": "questions are kept for thirty days. the notebook row holds nineteen fields, none of them about the reader. it does not hold an address, browser string, session, or cookie. the access log deletes address fields and writes REDACTED where the query string goes.", "digest_skipped": null }, { "id": "the-door-and-what-it-costs-to-keep-open", "heading": "The door, and what it costs to keep open", "level": 2, "span": [ 11598, 12759 ], "chunks": [ [ 11598, 12759 ] ], "chars": 1161, "digest": "the model is one seat, so there is a limit on how much any one reader can have. limits are 20 questions an hour, 40 a day, and no more than 6 in any one minute. limits are keyed per address, with a backstop per network block.", "digest_skipped": null }, { "id": "the-bench-what-we-measured-before-linking-it", "heading": "The bench: what we measured before linking it", "level": 2, "span": [ 12759, 18255 ], "chunks": [ [ 12759, 18255 ] ], "chars": 5496, "digest": "the assistant sat an exam of 62 questions drawn over 41 of the pages. the bench measures groundedness through verbatim figures, real citations, no redaction hits, and the abstain. the ninth run was taken on v1.0.1. the 95th-percentile seconds are 7.633 cold one in flight.", "digest_skipped": null }, { "id": "the-run-that-failed-and-what-it-changed", "heading": "The run that failed, and what it changed", "level": 2, "span": [ 18255, 19430 ], "chunks": [ [ 18255, 19430 ] ], "chars": 1175, "digest": "the seventh run failed the abstain floor on one arm. the model wrote two sentences when asked about a page's result compared to one published in 2028. a check struck the first sentence because 2028 is a name the article does not use.", "digest_skipped": null }, { "id": "what-it-got-wrong-on-the-way-here", "heading": "What it got wrong on the way here", "level": 2, "span": [ 19430, 22540 ], "chunks": [ [ 19430, 22540 ] ], "chars": 3110, "digest": "the assistant was audited in fifteen numbered rounds. findings included the model answering in lowercase, the draft showing citation syntax, and a receipt claiming a check that did not run. the service has 1,335 tests as this page is written at v1.1.1.", "digest_skipped": null }, { "id": "what-this-page-does-not-say", "heading": "What this page does not say", "level": 2, "span": [ 22540, 24540 ], "chunks": [ [ 22540, 24540 ] ], "chars": 2000, "digest": "this page makes no quality claim and no load claim beyond six in flight. the judge seat is empty and the uptime week is not yet recorded. being in beta means the door is narrow and the marks are the mechanism in public.", "digest_skipped": null }, { "id": "what-to-take-with-you", "heading": "What to take with you", "level": 2, "span": [ 24540, 25449 ], "chunks": [ [ 24540, 25449 ] ], "chars": 909, "digest": "the model reads the article up to about 82,800 tokens of prompt. buttons cost no model call. questions are kept for thirty days with no address, session, or cookie. the bench measures groundedness, not whether an answer is good.", "digest_skipped": null }, { "id": "how-to-check-our-work", "heading": "How to check our work", "level": 2, "span": [ 25449, 26514 ], "chunks": [ [ 25449, 26514 ] ], "chars": 1065, "digest": null, "digest_skipped": "credits" }, { "id": "who-ran-this-and-thanks", "heading": "Who ran this, and thanks", "level": 2, "span": [ 26514, 28205 ], "chunks": [ [ 26514, 28205 ] ], "chars": 1691, "digest": null, "digest_skipped": "credits" } ], "chips": [ { "id": "c-c315c647", "q": "how does the article room function?", "a": "Each article contains a unique room with a question box and buttons. A model reads the article to draft answers, which undergo six checks before being displayed. The room provides a short version, data found in numbers, suggested reading, and specific article sections.", "cites": [ "the-room-behind-every-article-and-what-it-opens" ], "published": false, "publish_state": "held", "quote": "", "quote_cite": "" }, { "id": "c-9f4e6206", "q": "what specific terms define the room's operation?", "a": "The operation relies on five key concepts: the Room, which is the question box and buttons; the Seat, the running model; the file beside the article, which contains summary and receipt data; the Fence, a marked block for text; and Abstain, the refusal to answer when information is missing.", "cites": [ "five-words-this-page-leans-on" ], "published": false, "publish_state": "held", "quote": "", "quote_cite": "" }, { "id": "c-e775b2f0", "q": "how are model-free answers generated?", "a": "Most information, such as summary lines and credits, is lifted directly from the author's words in the article. These are stored in a file beside the page, allowing buttons to provide immediate responses with zero model calls.", "cites": [ "where-the-answers-come-from" ], "published": false, "publish_state": "held", "quote": "", "quote_cite": "" }, { "id": "c-14c8fe8e", "q": "what happens if a draft sentence fails a check?", "a": "A sentence that fails a check is struck through, and the reason is provided in plain words, such as a figure not being in the article. If a sentence fails due to redaction or prohibited words, the words are removed entirely.", "cites": [ "what-happens-to-a-draft-before-you-see-it" ], "published": false, "publish_state": "held", "quote": "", "quote_cite": "" }, { "id": "c-addba695", "q": "what data is stored regarding user questions?", "a": "Questions are kept for thirty days to allow for human review. The recorded data includes the question, the page, the time, the answer's digest, cited sections, and various counters, but it excludes personal identifiers like addresses, cookies, or session IDs.", "cites": [ "what-it-keeps-and-what-it-never-writes-down" ], "published": false, "publish_state": "held", "quote": "", "quote_cite": "" }, { "id": "c-54d816ba", "q": "what limits exist for free-text questions?", "a": "Readers are limited to 20 questions an hour and 40 a day, with a restriction of no more than 6 in any one minute. These limits are keyed per address to prevent a single user or network block from exhausting the service.", "cites": [ "the-door-and-what-it-costs-to-keep-open" ], "published": false, "publish_state": "held", "quote": "", "quote_cite": "" }, { "id": "c-fff39449", "q": "what does the bench measurement verify?", "a": "The bench measures groundedness by verifying that every figure is verbatim, every citation is real, no redactions occur, and the model abstains when the page does not contain the answer. It does not measure the quality or goodness of the answers.", "cites": [ "the-bench-what-we-measured-before-linking-it" ], "published": false, "publish_state": "held", "quote": "", "quote_cite": "" }, { "id": "c-216f6053", "q": "what information is omitted from the article?", "a": "The article makes no quality claims, as the judge seat remains empty. It also does not provide information regarding the specific load beyond six in flight or comparisons to other model classes.", "cites": [ "what-this-page-does-not-say" ], "published": false, "publish_state": "held", "quote": "", "quote_cite": "" } ], "related": [ { "slug": "where-your-question-goes", "why": "shares ground with § What it keeps, and what it never writes down · § The first promise: your visit stays between us" }, { "slug": "how-we-work", "why": "shares ground with § What happens to a draft before you see it · § What to take with you" }, { "slug": "the-open-call", "why": "shares ground with § What happens to a draft before you see it · § Five things, before any figure" } ], "thanks": "## Who ran this, and thanks {#who-ran-this-and-thanks}\n\nThe corpus is this hub's own articles, most under CC BY 4.0 and the rest under the terms each page states; the room reads nothing else. The seat is a gemma-class open-weight writer from Google DeepMind. The licence file packaged with the fourth-generation weights reads Apache 2.0 and carries none of the custom terms earlier gemma generations packaged — a read of the redistributed blob, not of the vendor's own terms page, which is the distinction the hub's licence ledger draws and stamps. The weights this service loads are a quantised repack of it, and its licence is the repack's published one. The hub's licence ledger carries first-hand reads of the packaged blobs of the same family, and says in its own words where a read is a file and where it is a metadata tag. It is served by vLLM (Apache 2.0) with FP8 weights on PyTorch (BSD 3-clause) over CUDA, the one closed piece in the chain, named as such. The service is Python 3.14 standard library only, behind Caddy (Apache 2.0), managed by systemd (LGPL 2.1 or later); its notebook is a plain JSON-lines file. The rate limiter and the directive lint are our own code, vendored from another product of this workshop with a drift test. The related-page matrix beside each article was built once with a nomic-class embedding model (Apache 2.0) and committed; no model runs for it at request time. The bench harness, the packs and the twins are the hub's own tools.\n\nA small human team ran the benches, read the audits, and signed the numbers; a fleet of AI agents built, audited and re-audited the service under that team's rulings. None of the projects above owed us anything.", "kit": { "url": "https://research.strata2signal.com/ask-about-this-page/data/", "licence": "CC BY 4.0", "files": [ "NOTES.md", "PREREG.md", "README.md", "bank.json", "harness/packnorm.py", "harness/run.py", "provenance.json", "runs/r1-cold-c1.json", "runs/r1-cold-c6.json", "runs/r1-warm-c1.json", "runs/r1-warm-c6.json", "runs/r2-cold-c1.json", "runs/r2-cold-c6.json", "runs/r2-warm-c1.json", "runs/r2-warm-c6.json", "runs/r3-cold-c1.json", "runs/r3-cold-c6.json", "runs/r3-warm-c1.json", "runs/r3-warm-c6.json", "runs/r4-cold-c1.json", "runs/r4-cold-c6.json", "runs/r4-warm-c1.json", "runs/r4-warm-c6.json", "runs/r5-cold-c1.json", "runs/r5-cold-c6.json", "runs/r5-warm-c1.json", "runs/r5-warm-c6.json", "runs/r6-cold-c1.json", "runs/r6-cold-c6.json", "runs/r6-warm-c1.json", "runs/r6-warm-c6.json", "runs/r7-cold-c1.json", "runs/r7-cold-c6.json", "runs/r7-warm-c1.json", "runs/r7-warm-c6.json", "runs/r8-cold-c1.json", "runs/r8-cold-c1.log", "runs/r8-cold-c6.json", "runs/r8-cold-c6.log", "runs/r8-warm-c1.json", "runs/r8-warm-c1.log", "runs/r8-warm-c6.json", "runs/r8-warm-c6.log", "runs/r9-cold-c1.json", "runs/r9-cold-c1.log", "runs/r9-cold-c6.json", "runs/r9-cold-c6.log", "runs/r9-warm-c1.json", "runs/r9-warm-c1.log", "runs/r9-warm-c6.json", "runs/r9-warm-c6.log" ] }, "seat_class": { "writer": "gemma-class", "judge": null, "writer_runtime": "vllm" }, "bench": { "bullets_written": 3, "bullets_kept": 0, "digests_written": 11, "digests_kept": 11, "chips_written": 8, "chips_kept": 8, "chips_grounded": 0, "chips_published": 0, "dropped_by": { "new_noun": 1, "figure": 0, "length": 0, "cite": 0, "judge": 0, "quote": 0, "directive": 0, "redaction": 0, "profanity": 0, "identifier": 0, "hand-edit": 2 }, "new_noun_tokens_checked": [ "072", "1", "131", "20", "2028", "335", "40", "41", "6", "62", "7.633", "800", "82", "95th-percentile", "Abstain", "FP8", "Fence", "IDs", "REDACTED", "Room", "Seat", "v1.0.1", "v1.1.1" ], "new_noun_tokens_withheld": 1, "figure_definition": "v3", "drops": [ { "kind": "bullet", "reason": "new_noun", "item": null, "item_chars": 2, "withheld": "a string naming something the article does not", "detail": "" }, { "kind": "bullet", "reason": "hand-edit", "count": 2, "item": null, "withheld": "two bullets removed from this pack by hand", "detail": "removed from this pack by hand on 2026-09-15 at the release cut for exhibit forty-eight, after the writer ran and after the gates had counted. Both lines published a number without its unit, which is a bare number on a page: \"the model's window holds 131,072\" (the article's own words are \"a window of 131,072 tokens\") and \"the service has a total of 1,335\" (the article's own words are \"1,335 tests\"). Both figures are verbatim in their cited sections and gate 1 kept both; the unit is not something a gate on this shelf checks yet, and it is the same defect the forty-seven cut hand-dropped on 2026-09-15. The writer was re-run twice rather than edited, and both re-runs produced bare numbers too (\"the service answers about 1,500\", where the article says \"1,500 a day\"), so this is the writer's shape on this page and not one unlucky draw. Not a gate drop, and recorded here so the ledger adds up." } ] }, "generated_utc": "2026-09-15T18:01:19Z" }